Fire alarm system and method employing multi-layer net processing structure of detection value weight coefficients

Claim 2

A fire alarm system for monitoring the presence of a fire based on K fire related probabilities derived from the detection of I different fire related phenomena, where K and I are integers and I has a value of at least 2, comprising: sensor means for detecting said I different fire related phenomena and for generating I sensor detection values respectively indicative of the thus detected said I different fire related phenomena; a data table for storing in advance a plurality of combinations of I prestored detection values and a plurality of combinations of K prestored fire related probability values respectively associated with said plurality of combinations of I prestored detection values; a signal processing means, having a net structure formed of a plurality of network layers, for receiving said I sensor detection values from said sensor means and for generating K output fire related probability values respectively indicative of said K fire related probabalities, said plurality of network layers having weight coefficients assigned thereto, said signal processing means processing said I sensor detection values with said weight coefficients to obtain corresponding weighted detection values and processing said weighted detection values to obtain corresponding weighted intermediate detection values and processing said weighted intermediate detection values to obtain each of said K output fire related probability values; weight coefficient adjustment means for applying the plurality of combinations of I prestored detection values to said net structure of said signal processing means to arithmetically determine values of said weight coefficients to obtain corresponding K probability values output by said net structure for each applied combination of I prestored detection values which coincide with the associated combination of said K prestored fire related probability values stored in said data table.

Claim 3

A fire alarm system for monitoring the presence of a fire based on K fire related probabilities derived from the detection of I different fire related phenomena, where I and K are integers and I has a value of at least 2, comprising: sensor means for detecting said I different fire related phenomena and for generating I sensor detection values respectively indicative of the thus detected said I different fire related phenomena; a signal processing means, having a net structure formed of a plurality of network layers, for receiving said I sensor detection values from said sensor means and for generating K output fire related probability values respectively indicative of said K fire related probabilities, said plurality of network layers having weight coefficients assigned thereto, said signal processing means processing said I sensor detection values with said weight coefficients to obtain corresponding weighted detection values and processing said weighted detection values to obtain corresponding weighted intermediate detection values and processing said weighted intermediate detection values to obtain each of said K output fire related probability values; means for previously storing values of said weight coefficients determined by comparing a set of said K output fire related probability values obtained by applying a given set of I input sensor detection values to said signal processing means with a set of K target fire related probability values which are previously determined to be associated with said given set of I input sensor detection values, said weight coefficients are determined to minimize a difference between said set of said K output fire related probability values and said set of K target fire related probability values; wherein the thus stored values of said weight coefficients are said weight coefficients assigned to said plurality of network layers of said net structure.

Claim 4

A fire alarm system for monitoring the presence of a fire based on K fire related probabilities derived from the detection of I different fire related phenomena, where I and K are integers and I has a value of at least 2, comprising: sensor means for detecting said I different fire related phenomena and for generating I sensor detection values respectively indicative of the thus detected said I different fire related phenomena; a signal processing means, having a net structure formed of a plurality of network layers, for receiving said I sensor detection values from said sensor means and for generating K output fire related probability values respectively indicative of said K fire related probabilities, said plurality of network layers having weight coefficients assigned thereto, said signal processing means processing said I sensor detection values with said weight coefficients to obtain corresponding weighted detection values and processing said weighted detection values to obtain corresponding weighted intermediate detection values and processing said weighted intermediate detection values to obtain each of said K output fire related probability values; means for previously storing values of said weight coefficients determined by comparing a set of said K output fire related probability values obtained by applying a given set of I input sensor detection values to said signal processing means with a set of K target fire related probability values which are previously determined to be associated with said given set of I input sensor detection values, said weight coefficients are determined to minimize a difference between said set of said K output fire related probability values and said set of K target fire related probability values; wherein the thus stored values of said weight coefficients are said weight coefficients assigned to said plurality of network layers of said net structure.

Claim 5

A fire alarm system as recited in claim 1 or 2, said plurality of network layers of said net structure including I input network layers, J intermediate network layers and K output network layers each having said weight coefficients assigned thereto, each of said I input network layers receiving a corresponding one of said I sensor detection values to obtain said corresponding weighted detection values and outputting J weighted detection values using the weight coefficients assigned thereto, each of said J intermediate network layers receiving a corresponding one of said J weighted detection values output from each of the I input network layers, summing thus received I corresponding weighted detection values respectively output from said I input network layers to obtain said intermediate detection value, and outputting K weighted intermediate detection values by applying the weight coefficients assigned thereto to said intermediate detection value, and each of said K output network layers receiving a corresponding one of said K weighted intermediate detection values output from each of said J intermediate network layers and summing thus received J corresponding intermediate detection values respectively output from said J intermediate network layers to obtain and output a corresponding one of said K fire related probability values.

Claim 6

A fire alarm system as recited in claim 1 or 2, wherein said I different fire related phenomena include at least two of a smoke density, a gas concentration and temperature.

Claim 7

A fire alarm system as recited in claim 1 or 2, wherein said I different fire related phenomena include at least two of a smoke density, a gas concentration and temperature, and an environmental condition of a detected area.

Claim 8

A fire alarm system as recited in claim 2, further including a fire receiver and a plurality of fire detectors connected to said fire receiver, said sensor means located at said plurality of fire detectors, and said signal processing means and said storing means located at said fire receiver.

Claim 9

A fire alarm system as recited in claim 2, further including a fire receiver and a plurality of fire detectors connected to said fire receiver, said sensor means and said signal processing means and storing means located at said plurality of fire detectors.

Claim 10

In a fire alarm system for monitoring the presence of a fire based on one fire related probability derived from the detection of I different fire related phenomena, where I is an integer having a value of at least 2, a fire monitoring method comprising: (1) carrying out a learning process including the steps of (a) experimentally obtaining a plurality of combinations of I detection values and a plurality of probability values respectively associated with said plurality of combinations of I detection values, (b) temporarily assigning values to first weight coefficients applied to said plurality of combinations of I detection values, (c) applying the first weight coefficients assigned in step (b) to said I detection values of a first one of said plurality of combinations of I detection values and summing the thus weighted I detection values to obtain a first sum value for each of J intermediate detection values, (d) temporarily assigning values to second weight coefficients applied to said J intermediate detection values, (e) applying the second weight coefficients assigned in step (d) to said J intermediate detection values obtained by applying the first weight coefficients to said I detection values of said first one of said plurality of combinations of I detection values and summing the thus weighted J intermediate detection values to obtain a second sum value for a probability value, (f) comparing said second sum value with one of said probability values associated with said first one of said plurality of combinations of I detection values to obtain a compared value, (g) repeating said steps (c), (e) and (f) with respect to each of the remaining said plurality of combinations of I detection values in succession, and summing absolute values of a plurality of said compared values obtained in said step (f) to obtain an absolute sum, (h) continuously repeating said steps (b) through (g) by varying values of said first and said second weight coefficients until said values of said first and said second weight coefficients are obtained which minimize said absolute sum, and (i) storing said values of said first and said second weight coefficients obtained in said step (h) which minimize said absolute sum; and (2) carrying out a detection process including the steps of (a) sensing I fire related phenomena and generating corresponding I sensor detection values, (b) applying the first weight coefficients stored in said step (i) of said learning process to the I sensor detection values and summing the thus weighted I sensor detection values to obtain J intermediate sensor detection values, (c) applying the second weight coefficients stored in said step (i) of said learning process to said J intermediate sensor detection values and summing the thus weighted J intermediate sensor detection values to obtain one fire probability value, and (d) determining the probability of a fire based on the one fire probability value.

Claim 11

In a fire alarm system for monitoring the presence of a fire based on one fire related probability derived from the detection of I different fire related phenomena, where I is an integer having a value of at least 2, a fire monitoring method comprising: (1) carrying out a learning process including the steps of (a) experimentally obtaining a plurality of combinations of I detection values and a plurality of probability values respectively associated with said plurality of combinations of I detection values, (b) temporarily assigning values to first weight coefficients applied to said plurality of combinations of I detection values, (c) applying the first weight coefficients assigned in step (b) to said I detection values of a first one of said plurality of combinations of I detection values and summing the thus weighted I detection values to obtain a first sum value for each of J intermediate detection values, (d) temporarily assigning values to second weight coefficients applied to said J intermediate detection values, (e) applying the second weight coefficients assigned in step (d) to said J intermediate detection values obtained by applying the first weight coefficients to said I detection values of said first one of said plurality of combinations of I detection values and summing the thus weighted J intermediate detection values to obtain a second sum value for a probability value, (f) comparing said second sum value with one of said probability values associated with said first one of said plurality of combinations of I detection values to obtain a compared value, (g) repeating said steps (c), (e) and (f) with respect to each of the remaining said plurality of combinations of I detection values in succession, and summing absolute values of a plurality of said compared values obtained in said step (f) to obtain an absolute sum, (h) continuously repeating said steps (b) through (g) by varying values of said first and said second weight coefficients until said values of said first and said second weight coefficients are obtained which minimize said absolute sum, and (i) storing said values of said first and said second weight coefficients obtained in said step (h) which minimize said absolute sum; and (2) carrying out a detection process including the steps of (a) sensing I fire related phenomena and generating corresponding I sensor detection values, (b) applying the first weight coefficients stored in said step (i) of said learning process to the I sensor detection values and summing the thus weighted I sensor detection values to obtain J intermediate sensor detection values, (c) applying the second weight coefficients stored in said step (i) of said learning process to said J intermediate sensor detection values and summing the thus weighted J intermediate sensor detection values to obtain one fire probability value, and (d) determining the probability of a fire based on the one fire probability value.

Claim 12

In a fire alarm system for monitoring the presence of a fire based on K fire related probabilities derived from the detection of I different fire related phenomena, where I and K are integers having a value of at least 2, a fire monitoring method comprising: (1) carrying out a learning process including the steps of (a) experimentally obtaining a plurality of combinations of I detection values and a plurality of combinations of K probability values respectively associated with said plurality of combinations of I detection values, (b) temporarily assigning values to first weight coefficients applied to said plurality of combinations of I detection values, (c) applying the first weight coefficients assigned in step (b) to said I detection values of a first one of said plurality of combinations of I detection values and summing the thus weighted I detection values to obtain a first sum value for each of J intermediate detection values, (d) temporarily assigning values to second weight coefficients applied to said J intermediate detection values, (e) applying the second weight coefficients assigned in step (d) to said J intermediate detection values obtained by applying the first weight coefficients to said I detection values of said first one of said plurality of combinations of I detection values and summing the thus weighted J intermediate detection values to obtain a second sum value for each of K probability values, (f) comparing said K second sum values respectively with the K probability values of a one of said plurality of combinations of said K probability values associated with said first one of said plurality of combinations of I detection values to obtain a compared value for each of said K second sum values and summing absolute values of the thus obtained K compared values to obtain an absolute sum, (g) repeating said steps (c), (e) and (f) with respect to each of the remaining said plurality of combinations of I detection values in succession, and summing the absolute sums obtained in said step (f) to obtain a total absolute sum, (h) continuously repeating said steps (b) through (g) by varying values of said first and said second weight coefficients until said values of said first and said second weight coefficients are obtained which minimize said total absolute sum, and (i) storing said values of said first and second weight coefficients obtained in said step (h) which minimize said total absolute sum; and (2) carrying out a detection process including the steps of (a) sensing I fire related phenomena and generating corresponding I sensor detection values, (b) applying the first weight coefficients stored in said step (i) of said learning process to the I sensor detection values and summing the thus weighted I sensor detection values to obtain J intermediate sensor detection values, (c) applying the second weight coefficients stored in said step (i) of said learning process to said J intermediate sensor detection values and summing the thus weighted J intermediate sensor detection values to obtain said K fire probabilities, and (d) determining the probability of a fire based on said K fire probabilities.

Claim 13

In a fire alarm system for monitoring the presence of a fire based on K fire related probabilities derived from the detection of I different fire related phenomena, where I and K are integers having a value of at least 2, a fire monitoring method comprising: (1) carrying out a learning process including the steps of (a) experimentally obtaining a plurality of combinations of I detection values and a plurality of combinations of K probability values respectively associated with said plurality of combinations of I detection values, (b) temporarily assigning values to first weight coefficients applied to said plurality of combinations of I detection values, (c) applying the first weight coefficients assigned in step (b) to said I detection values of a first one of said plurality of combinations of I detection values and summing the thus weighted I detection values to obtain a first sum value for each of J intermediate detection values, (d) temporarily assigning values to second weight coefficients applied to said J intermediate detection values, (e) applying the second weight coefficients assigned in step (d) to said J intermediate detection values obtained by applying the first weight coefficients to said I detection values of said first one of said plurality of combinations of I detection values and summing the thus weighted J intermediate detection values to obtain a second sum value for each of K probability values, (f) comparing said K second sum values respectively with the K probability values of a one of said plurality of combinations of said K probability values associated with said first one of said plurality of combinations of I detection values to obtain a compared value for each of said K second sum values and summing absolute values of the thus obtained K compared values to obtain an absolute sum, (g) repeating said steps (c), (e) and (f) with respect to each of the remaining said plurality of combinations of I detection values in succession, and summing the absolute sums obtained in said step (f) to obtain a total absolute sum, (h) continuously repeating said steps (b) through (g) by varying values of said first and said second weight coefficients until said values of said first and said second weight coefficients are obtained which minimize said total absolute sum, and (i) storing said values of said first and second weight coefficients obtained in said step (h) which minimize said total absolute sum; and (2) carrying out a detection process including the steps of (a) sensing I fire related phenomena and generating corresponding I sensor detection values, (b) applying the first weight coefficients stored in said step (i) of said learning process to the I sensor detection values and summing the thus weighted I sensor detection values to obtain J intermediate sensor detection values, (c) applying the second weight coefficients stored in said step (i) of said learning process to said J intermediate sensor detection values and summing the thus weighted J intermediate sensor detection values to obtain said K fire probabilities, and (d) determining the probability of a fire based on said K fire probabilities.

Claim 14

A fire monitoring method as recited in claim 9, wherein said learning process is carried out as part of an initialization of said fire alarm system after said fire alarm system has been installed at an operating location.

Claim 15

A fire monitoring method as recited in claim 9, wherein said learning process is carried during the course of manufacturing said fire alarm system.